@@ -3,12 +3,20 @@ import { columnTypeOf } from '@/lib/table/column-types/registry'
 import { validatedTimestampSql } from '@/lib/table/column-types/timestamp-sql'
 import type { ColumnDefinition } from '@/lib/table/types'
 
+/**
+ * The database cast an equality match on `column` compares through, or `undefined` when equality is
+ * JSONB containment. Only a type with an equality projection compares by cast, which no index serves.
+ */
+export function equalityCast(column: ColumnDefinition): string | undefined {
+  const definition = columnTypeOf(column)
+  return definition.valueForEquality ? (definition.jsonbCast ?? undefined) : undefined
+}
+
 /** Casts stored text for equality, treating malformed explicit timestamps as absent instants. */
 export function columnTextForEquality(cell: SQL, column: ColumnDefinition): SQL {
-  const definition = columnTypeOf(column)
-  if (!definition.valueForEquality || !definition.jsonbCast) return cell
-  if (definition.timestampValidation) {
-    return validatedTimestampSql(cell, definition.timestampValidation)
-  }
-  return sql`(${cell})::${sql.raw(definition.jsonbCast)}`
+  const cast = equalityCast(column)
+  if (!cast) return cell
+  const { timestampValidation } = columnTypeOf(column)
+  if (timestampValidation) return validatedTimestampSql(cell, timestampValidation)
+  return sql`(${cell})::${sql.raw(cast)}`
 }
@@ -44,6 +44,7 @@ import { lockUniqueColumns, lockUniqueValues } from '@/lib/table/rows/unique-loc
 import { getTableById } from '@/lib/table/service'
 import { getOrCreateTableSnapshot } from '@/lib/table/snapshot-cache'
 import type { ColumnDefinition, JsonValue, RowData, TableDefinition } from '@/lib/table/types'
+import { checkBatchUniqueConstraintsDb } from '@/lib/table/validation'
 
 const url = readTestDatabaseUrl()
 if (process.env.DATABASE_URL !== url) {
@@ -87,7 +88,7 @@ async function createTable(columns: ColumnDefinition[]): Promise<TableDefinition
 
 async function seedRows(
   tableId: string,
-  rows: Array<{ id: string; data: RowData; orderKey: string | null }>
+  rows: Array<{ id: string; data: RowData; orderKey: string | null; position?: number }>
 ) {
   await db.insert(userTableRows).values(rows.map((row) => ({ ...row, tableId, workspaceId })))
 }
@@ -481,6 +482,33 @@ describe('table row writes against real PostgreSQL', () => {
       expect(await storedCount(table.id, { email: 'dup@example.test' })).toBe(1)
     })
 
+    it('rejects the second of two concurrent batches over the value-lock cap that share a value', async () => {
+      const table = await createTable(uniqueColumns)
+      const batch = (prefix: string) => () =>
+        batchInsertRows(
+          {
+            tableId: table.id,
+            workspaceId,
+            rows: [
+              ...Array.from({ length: 100 }, (_, i) => ({ email: `${prefix}-${i}@example.test` })),
+              { email: 'shared@example.test' },
+            ],
+            secretProvenance: undefined,
+            capabilityGovernedUserId: null,
+          },
+          table,
+          'unique-race'
+        )
+
+      const results = await raceUnderHeldDefinitionRow(table.id, [batch('left'), batch('right')], {
+        onDefinitionRow: 1,
+        onValueLock: 1,
+      })
+
+      expect(fulfilled(results)).toBe(1)
+      expect(await storedCount(table.id, { email: 'shared@example.test' })).toBe(1)
+    })
+
     it('rejects an import batch racing a single insert of the same value', async () => {
       const table = await createTable(uniqueColumns)
 
@@ -700,6 +728,163 @@ describe('table row writes against real PostgreSQL', () => {
     })
   })
 
+  describe('batch unique check across many values', () => {
+    /** Seeds `cells` at positions 0.., which the conflict messages report one-based. */
+    const seedAt = (tableId: string, cells: RowData[]) =>
+      seedRows(
+        tableId,
+        cells.map((data, position) => ({
+          id: `${tableId}-${position}`,
+          data,
+          orderKey: `a${position}`,
+          position,
+        }))
+      )
+
+    /**
+     * Index and sequential scans this backend has counted on `user_table_rows`, by relation. The
+     * counters are the backend's unflushed statistics, which can still hold scans from earlier
+     * transactions on the same pooled connection, so callers diff two readings.
+     */
+    async function rowScans(trx: DbTransaction): Promise<Map<string, number>> {
+      const scans = await trx.execute<{ relname: string; scans: number }>(sql`
+        SELECT c.relname, pg_stat_get_xact_numscans(c.oid)::int AS scans
+        FROM pg_class c
+        WHERE c.oid = 'user_table_rows'::regclass
+          OR c.oid IN (SELECT indexrelid FROM pg_index WHERE indrelid = 'user_table_rows'::regclass)`)
+      return new Map(scans.map((row) => [row.relname, row.scans]))
+    }
+
+    /** Runs the check in a transaction and returns its result with the scans the check alone ran. */
+    const checkWithScans = (table: TableDefinition, rows: RowData[]) =>
+      db.transaction(async (trx) => {
+        // A parallel worker's scans are not counted in the leader's transaction statistics.
+        await trx.execute(sql`SET LOCAL max_parallel_workers_per_gather = 0`)
+        const before = await rowScans(trx)
+        const result = await checkBatchUniqueConstraintsDb(table.id, rows, table.schema, trx)
+        const scans: Record<string, number> = {}
+        for (const [relname, count] of await rowScans(trx)) {
+          const ran = count - (before.get(relname) ?? 0)
+          if (ran > 0) scans[relname] = ran
+        }
+        return { result, scans }
+      })
+
+    it('reports stored and in-batch conflicts across every probe group', async () => {
+      const table = await createTable([
+        { id: 'email', name: 'email', type: 'string', unique: true },
+        { id: 'meta', name: 'meta', type: 'json', unique: true },
+      ])
+      await seedAt(table.id, [
+        { email: 'first@example.test' },
+        { email: 'middle@example.test' },
+        { email: 'last@example.test' },
+        { meta: { a: 1, b: 2 } },
+        { meta: [1, 2] },
+      ])
+      const rows: RowData[] = Array.from({ length: 260 }, (_, i) => ({
+        email: `new-${i}@example.test`,
+      }))
+      rows[0] = { email: 'new-0@example.test', meta: { b: 2, a: 1 } }
+      rows[1] = { email: 'new-1@example.test', meta: { a: 1 } }
+      rows[2] = { email: 'new-2@example.test', meta: [2, 1] }
+      rows[3] = { email: 'first@example.test', meta: [1, 2] }
+      rows[7] = { email: null }
+      rows[8] = {}
+      rows[130] = { email: 'middle@example.test' }
+      rows[200] = { email: 'new-5@example.test' }
+      rows[259] = { email: 'last@example.test' }
+
+      expect((await checkWithScans(table, rows)).result).toEqual({
+        valid: false,
+        errors: [
+          {
+            row: 0,
+            errors: [
+              'Column "meta" must be unique. Value "[object Object]" already exists in row 4',
+            ],
+          },
+          {
+            row: 3,
+            errors: [
+              'Column "email" must be unique. Value "first@example.test" already exists in row 1',
+              'Column "meta" must be unique. Value "1,2" already exists in row 5',
+            ],
+          },
+          {
+            row: 130,
+            errors: [
+              'Column "email" must be unique. Value "middle@example.test" already exists in row 2',
+            ],
+          },
+          {
+            row: 200,
+            errors: [
+              'Column "email" must be unique. Value "new-5@example.test" duplicates row 6 in batch',
+            ],
+          },
+          {
+            row: 259,
+            errors: [
+              'Column "email" must be unique. Value "last@example.test" already exists in row 3',
+            ],
+          },
+        ],
+      })
+    })
+
+    it('matches an expiration by instant in one pass over the table', async () => {
+      const table = await createTable([
+        { id: 'expires', name: 'expires', type: 'ttl', unique: true },
+      ])
+      await seedAt(table.id, [{ expires: '2024-01-31T00:00:00-00:00' }])
+      const rows: RowData[] = Array.from({ length: 30 }, (_, i) => ({
+        expires: `2025-01-${String(i + 1).padStart(2, '0')}T00:00:00-00:00`,
+      }))
+      rows[27] = { expires: '2024-01-31T01:00:00+01:00' }
+      rows[28] = { expires: null }
+
+      const { result, scans } = await checkWithScans(table, rows)
+
+      expect(result).toEqual({
+        valid: false,
+        errors: [
+          {
+            row: 27,
+            errors: [
+              'Column "expires" must be unique. Value "2024-01-31T01:00:00+01:00" already exists in row 1',
+            ],
+          },
+        ],
+      })
+      expect(Object.values(scans).reduce((sum, count) => sum + count, 0)).toBe(1)
+    })
+
+    it('probes the containment index once per value on a large table', async () => {
+      const table = await createTable([
+        { id: 'email', name: 'email', type: 'string', unique: true },
+      ])
+      await control`INSERT INTO user_table_rows (id, table_id, workspace_id, data, position, order_key)
+        SELECT ${table.id}::text || '-' || i, ${table.id}, ${workspaceId},
+          jsonb_build_object('email', 'stored-' || i || '@example.test'), i, 'a' || i
+        FROM generate_series(0, 39999) AS i`
+      // A bulk insert leaves its entries in the GIN pending list, which a manual ANALYZE does not
+      // merge and which the planner charges to every probe, so flush it for a deterministic plan.
+      await control`SELECT gin_clean_pending_list('user_table_rows_tenant_data_gin_idx'::regclass)`
+      await control`ANALYZE user_table_rows`
+      const rows: RowData[] = Array.from({ length: 260 }, (_, i) => ({
+        email: i % 10 === 0 ? `stored-${i * 100}@example.test` : `new-${i}@example.test`,
+      }))
+
+      const { result, scans } = await checkWithScans(table, rows)
+
+      await control`DELETE FROM user_table_rows WHERE table_id = ${table.id}`
+
+      expect(result.errors).toHaveLength(26)
+      expect(scans).toEqual({ user_table_rows_tenant_data_gin_idx: 260 })
+    }, 60_000)
+  })
+
   describe('writers validate against the live schema, not their snapshot', () => {
     const columns: ColumnDefinition[] = [
       { id: 'key', name: 'key', type: 'string', unique: true },
 
@@ -19,7 +19,7 @@ import {
   SINGLE_SELECT_OPERATORS,
   SINGLE_SELECT_OPS,
 } from '@/lib/table/column-types'
-import { columnTextForEquality } from '@/lib/table/column-types/comparison-sql'
+import { columnTextForEquality, equalityCast } from '@/lib/table/column-types/comparison-sql'
 import { NAME_PATTERN } from '@/lib/table/constants'
 import { normalizeDateCellValue } from '@/lib/table/dates'
 import { TableQueryValidationError } from '@/lib/table/errors'
@@ -630,10 +630,10 @@ export function fieldPredicate(
       : value
 
   const equalityClause = (operand: JsonValue): SQL => {
-    const definition = column && columnTypeOf(column)
-    if (column && operand !== null && definition?.valueForEquality && definition.jsonbCast) {
+    const cast = column && equalityCast(column)
+    if (column && operand !== null && cast) {
       const cell = columnTextForEquality(sql.raw(`${tableName}.data->>'${field}'`), column)
-      return sql`COALESCE(${cell} = ${operand}::${sql.raw(definition.jsonbCast)}, false)`
+      return sql`COALESCE(${cell} = ${operand}::${sql.raw(cast)}, false)`
     }
     return buildContainmentClause(tableName, field, operand)
   }
 
@@ -4,6 +4,7 @@
 
 import { db } from '@sim/db'
 import { userTableRows } from '@sim/db/schema'
+import { chunkArray } from '@sim/utils/helpers'
 import { and, eq, or, type SQL, sql } from 'drizzle-orm'
 import { NextResponse } from 'next/server'
 import { canonicalJson } from '@/lib/api/cursor-binding'
@@ -19,6 +20,7 @@ import {
   validateColumnTypeLimits,
   validateTypeMetadata,
 } from '@/lib/table/column-types'
+import { equalityCast } from '@/lib/table/column-types/comparison-sql'
 import {
   getMaxRowSizeBytes,
   NAME_PATTERN,
@@ -537,6 +539,13 @@ async function withSeqscanOffOn(
  */
 type UniqueCheckExecutor = Pick<typeof db, 'select' | 'execute'>
 
+/**
+ * Most values one branch of the batch conflict query ORs together. The branches join with UNION ALL
+ * and carry no LIMIT, so each value stays a GIN probe: under a LIMIT the planner walks every row of
+ * the table instead, and a much larger OR adds a scan of the table's whole entry in that index.
+ */
+const UNIQUE_PROBE_GROUP_SIZE = 25
+
 /**
  * Checks unique constraints for a batch of rows using targeted database queries.
  * Validates both against existing database rows and within the batch itself.
@@ -615,8 +624,7 @@ export async function checkBatchUniqueConstraintsDb(
         throw new Error(`Invalid column id: ${columnId}`)
       }
 
-      const valueArray = Array.from(values)
-      const valueConditions = valueArray.map((normalizedValue) => {
+      const valueConditions = Array.from(values, (normalizedValue) => {
         // Reconstruct the original typed value from its normalized key. Both
         // directions go through JSON unconditionally: keying the write on the
         // value's RUNTIME type while keying the read on the column's DECLARED
@@ -627,20 +635,26 @@ export async function checkBatchUniqueConstraintsDb(
         return uniqueValuePredicate(USER_TABLE_ROWS_SQL_NAME, columnId, originalValue, column)
       })
 
-      const conflictingRows = await ex
-        .select({
-          id: userTableRows.id,
-          data: userTableRows.data,
-          position: userTableRows.position,
-        })
-        .from(userTableRows)
-        .where(and(eq(userTableRows.tableId, tableId), or(...valueConditions)))
-        .limit(valueArray.length) // We only need up to one conflict per value
+      // A cast comparison is a filter over the table's rows whatever its shape, so it stays one
+      // pass rather than one per group.
+      const groups = equalityCast(column)
+        ? [valueConditions]
+        : chunkArray(valueConditions, UNIQUE_PROBE_GROUP_SIZE)
+      const conflictingRows = await ex.execute<{ data: RowData; position: number }>(
+        sql.join(
+          groups.map((group) =>
+            ex
+              .select({ data: userTableRows.data, position: userTableRows.position })
+              .from(userTableRows)
+              .where(and(eq(userTableRows.tableId, tableId), or(...group)))
+          ),
+          sql` UNION ALL `
+        )
+      )
 
       // Map conflicts back to batch rows
       for (const conflict of conflictingRows) {
-        const conflictData = conflict.data as RowData
-        const conflictValue = cellOf(conflictData, columnId)
+        const conflictValue = cellOf(conflict.data, columnId)
         if (conflictValue === null || conflictValue === undefined) continue
         // Keyed like the batch, since stored jsonb comes back with its keys reordered.
         const normalizedConflictValue = uniqueValueKey(conflictValue, column)